Giter Club home page Giter Club logo

llm_tool_using_collections's Introduction

LLM Tool using Collections

by WangYC @ Zhipu.AI

[toc]

Toolformer

https://arxiv.org/abs/2302.04761

Meta AI

Feb 2023

Tools

格式统一为<tool名称(输入)>的格式

Area Tool
QA Atlas Model
Calculator 自己写的加减乘除法函数 精确到小数点后两位
Web Search BM25 retriever
Machine Translation NLLB Model
Calendar calendar API

contrubutions

没明确写 自己总结:

  • 第一篇tool using的paper
  • zero-shot的模型推理能力(相较于后来的)

methods

bootstrapping 的过程中采用了特殊的loss定义来保证api的调用位置以及调用格式是符合要求的

同时在bootstrapping的过程中采用了heuristics的方法保证数据集的利用效率

image-20230331140256804

finetune过程是语言建模任务finetune的GPT-J

evaluation

数据集是CCNet的subset

baselines:

image-20230331140454693

downstream tasks

Area Task
General LAMA subset
Math AsSDiv SVAMP MAWPS
QA WebQS NQ TriviaQA
MultiLingual QA MLQA
Temporal Dataset TEMPLAMA
Language Modeling WikiText CCNet

weakness

evaluation的过程是分别打的不同area的榜,比如QA的榜、math的榜分开,而没有打一个综合的榜,不涉及不同Area的api的嵌套使用

ART

Tools

Area Tool
Search SerpAPI(Google Search)
Code generation Codex
Code Exicution virtual Python environment

contributions

image-20230331144759494

methods

筛选tool的思路

https://github.com/google/BIG-bench/blob/main/bigbench/benchmark_tasks/keywords_to_tasks.md#summary-table

从bigbench中提炼出的最常用skills:

Visual Reasoning, Temporal Reasoning, Propositional logic, Natural Logic, Machine Translation, Web Search, Knowledge Base or Database lookup, Recursive sub-question decomposition, Long text understanding, Database Operations, Algebra and Arithmetic, Code Generation and Editing, Text Tagging/Annotation(linguistic markers), Specialized Search(eg. looking up linguistic knowledge, scientific knowledge etc), String editing, Recursive operations over multiple choices, Topic classification, Evidence extraction, conditional Text Generation/Editing, and Sentence similarity.

随后选取了most used的5个task:

image-20230331154836002

task choosing的过程

https://github.com/bhargaviparanjape/language-programmes/blob/master/src/affordance/tasks/prompt_library.py

![image-20230331154523263](LLM Tool using Collections.assets/image-20230331154523263.png)

![image-20230331154356255](LLM Tool using Collections.assets/image-20230331154356255.png)

evaluation

Input-Output Instances of Big bench:

mainly text classification and short answer generation tasks in English

baselines

![image-20230331164205858](LLM Tool using Collections.assets/image-20230331164205858.png)

downstream tasks

![image-20230331164220140](LLM Tool using Collections.assets/image-20230331164220140.png)

weakness

执行是用python code执行的,并不是LLM执行。

通用目标。

Party

https://github.com/UT-MPC/homegpt

Tools

硬件灯

contributions

sementic parsing统一为json格式

高级抽象化的natural language的query parsing

methods

prmopt engineering

image-20230331165732353

evaluation

baselines

downstream tasks

weakness

没有采用LM执行

家具的数量有限,状态易于表示,相当于是一个状态机

Matrix.AI

HuggingGPT

llm_tool_using_collections's People

Contributors

wangyc-99 avatar

Watchers

 avatar

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.